Nearly half of small firms now use AI, but fewer than one in five feel confident: close the gap in one team hour, not a new hire

By Jason Sibley

Two recent surveys point the same way. Many small firms use AI, but far fewer feel confident using it. The usual instinct is to find someone who "gets AI" and hand it over. I'd try something cheaper first: one hour with the whole team and a shared list of dos and don'ts.

What the surveys say

Simply Business is a business insurer. Its 2026 SME Insights Report is its own customer-facing survey, so read it with that in mind. Published on 8 September 2026, it surveyed 1,842 UK small business owners between 30 July and 7 August 2026. Simply Business reports that:

• 47% use AI tools, up from 22% in 2025
• 19% are "very confident" using AI day to day
• 33% use it only for routine admin
• 31% don't understand how to use it, or are wary of bringing it in
• the main barriers are security and privacy (44%), no clear use (39%) and accuracy (36%)

Its chief executive, Julie Fisher, says small businesses need "tailored guidance on how AI can be used, accessible tools, and time to discover how it can work for them on their terms."

The Social Market Foundation's report, Byte-sized growth, came out on 19 October 2026. It surveyed 1,015 UK SME leaders and found that 60% of micro firms use AI, against 86% of small firms and 92% of medium ones. It also says advanced users report significantly better impacts than basic users. HR, sales, supply chain and IT appear to be the highest-value applications. Its view is that effective support "cannot rely on financial incentives alone".

The surveys measure usage differently, which is why the headline numbers don't match. I wouldn't add them together or average them. Both suggest that AI use is common, while confidence and depth of use lag behind.

Why confidence, not access, is the gap

The barriers people give are security and privacy, no clear use, and accuracy. None is "I can't get hold of the tool".

Each concerns behaviour that a team can agree on together:

• Address security worries with a plain rule on what data never goes in.
• Address "No clear use" by trying the tool on real tasks from your week.
• Address accuracy worries by checking every figure.

A hire doesn't fix that. One person gets confident, and everyone else either queues behind them or quietly carries on using their own free accounts. I describe that second pattern in shadow AI in small teams. An "AI person" may be the right hire for other reasons. But the confidence gap spans the whole team, and an agreed rule closes more of it than one expert's private skill.

The Social Market Foundation doesn't say how people become advanced users, so I'm not claiming a session does it. Practising on real work together is a place to start. If you'd rather build your own habit alone first, see one AI workflow for 30 days.

The one-hour session

Pick a quiet hour and ask everyone to bring a laptop. Use one agreed tool, ideally one your business already pays for. For a wider view of what a team can get done, see do more with the team you have.

Agenda

• 10 minutes: the do/don't list. Cover what data never goes in, why every figure gets checked, and who people ask when unsure.
• 20 minutes: everyone brings one real task from their week and tries it live. A customer email, a rota, a quote summary, whatever is sitting on their desk. The ChatGPT prompts guide has starting points if you need help with wording.
• 15 minutes: compare results. Write down what worked, what came out wrong and what took longer than doing it by hand.
• 10 minutes: agree two tasks to use AI for this month. Only two. Small and boring is fine.
• 5 minutes: decide who owns the list. That person keeps it current and answers "can I put this in?"

A starter do/don't list

This is a starting point, not legal advice. For questions about data protection or customer and staff information, speak to a solicitor. For a fuller version, see a simple AI policy for small teams and whether ChatGPT is safe for business data.

• Don't paste customer or staff personal data into a free tool.
• Don't paste anything confidential, such as agreed pricing or contracts, until the team has decided what's allowed.
• Do check every number, name and date before it goes anywhere.
• Do say when AI drafted something customer-facing, if your policy says so.
• Don't let AI send anything. A person reads it and a person sends it.
• Do ask the list owner before guessing.
• Do tell the team when you catch a mistake. That's useful information, not an embarrassment.

If you'd like a draft to edit before the session, you could paste this into your AI tool:

I run a small UK business. Please draft a one-page "do and don't" list for my team's use of AI tools, in plain British English.

Cover what data must never be pasted in, checking figures and names, who to ask when unsure, and whether to say when AI drafted something customer-facing.

Do not invent facts, laws or company policies. Where you would need information from me, write [OWNER TO FILL IN] instead of guessing.

Keep it under 300 words.

Read the result with your team and cut anything that doesn't fit how you work.

What to measure after a month

A short check-in at the end of the month is enough, using three simple counts:

• Tasks used: how many times the two agreed tasks were done with AI, and whether anyone added a third.
• Time saved: each person's rough estimate, labelled as an estimate. Don't dress a guess up as a figure.
• Mistakes caught: how many errors the checking habit spotted before they went out. A few is a good sign, because it shows people are checking.

Then ask the team whether they feel more or less sure than they did a month ago. For more detail, see measuring AI ROI for a small business.

If you'd like a hand

At My AI Apprentice we run AI sessions for small businesses, including this kind of team hour, if you'd rather not run it yourself. No pressure either way.